Experimental study on stacked autoencoder on insufficient training samples

Abstract

Many studies have shown that deep learning outperforms traditional machine learning methods in many applications. To prevent overfitting, a huge number of training samples is usually required in training process of deep learning. However, collecting such a large dataset is time-consumed and costly. Recently, several methods have been proposed to effectively learn the models with a limited number of training samples. In this paper, we experimentally evaluate the data augmentation and the transfer learning methods on the MNIST dataset with stacked autoencoders. The experiment results suggest that the transfer learning method can improve the performance of the SAE when training samples are insufficient.

Lab Members

Patrick Chan

Associate Professor, Vice Dean

Patrick Chan works on machine learning, deep learning, image processing, adversarial learning, and secure machine learning.